You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The code for ResNet model in the official/resnet preprocesses images with pixel values in the range of [0, 255] and subtracts the corresponding mean. The pretrained weights linked in the Readme of official/resnet are however for a model that expects the pixel values to be additionally donwscaled by 255. This problem is present in the master branch and the 1.7 branch.
It seems that in the 1.6 branch of tensorflow/models, things are even more confusing: there the subtracted means are divided by 255, but as far as I can see, the images are not.